Mallard 0.8.0
Unstructured, GPU accelerated, high-order finite volume CFD solver
Mallard website
Loading...
Searching...
No Matches
device_comm.h File Reference

Stream-ordered communication of device buffers (NCCL), and ordering between execution space instances without host synchronization. More...

#include <Kokkos_Core.hpp>

Go to the source code of this file.

Namespaces

namespace  comm
 

Functions

constexpr bool comm::nccl_available ()
 Whether this build can exchange device buffers with NCCL.
 
void comm::enqueue_wait (const Kokkos::DefaultExecutionSpace &waiter, const Kokkos::DefaultExecutionSpace &done)
 Make the work enqueued later on waiter wait for the work enqueued so far on done, without blocking the host where the backend allows it (CUDA); a fence of done otherwise.
 

Detailed Description

Stream-ordered communication of device buffers (NCCL), and ordering between execution space instances without host synchronization.

Author
Matthew Bonanni (mbona.nosp@m.nni0.nosp@m.01@gm.nosp@m.ail..nosp@m.com)
Version
0.1
Date
2026-10-05